ENAEpistemic Network Analysis Hub of Knowledge
Back to ENA Academy
InterpretationIntermediateAug 9, 202625 minENA Academy Lesson 04

Interpret ENA Networks With Qualitative Evidence

Read points, edges, and comparison networks as coordinated model outputs, then return to coded excerpts before writing a bounded claim.

Method sources

Shaffer, Collier, and Ruis: A Tutorial on Epistemic Network Analysis Epistemic Analytics: official ENA resources Shaffer and Ruis: ENA worked example of theory-based learning analytics
network interpretationdifference graphsqualitative evidenceevidence boundaries

Full tutorial

An ENA plot is an entry point into the evidence, not a self-interpreting picture. A point locates a unit or group network in a shared projected space. A network graph shows the relative strengths of connections among codes. A comparison network emphasizes which connections are relatively stronger in one group than another. These views answer related questions, but they are not interchangeable.

This tutorial completes the synthetic teacher-design example used across the ENA Academy pathway. You will move from a visible difference to the line-level records that produced it, test alternative explanations, and draft a claim whose strength matches the teaching dataset.

Teaching case

The same synthetic design-talk study

Suppose the scaffolded teams show a relatively stronger Evidence-Revision connection than the baseline teams. Your task is not to celebrate the thicker edge. Your task is to determine what the edge represents, which units contribute to it, which excerpts support it, and what the small fictional design cannot establish.

Practice dataset

Synthetic design-talk datasetA fictional 48-row CSV for learning the workflow, not for substantive inference.Codebook and coding examplesDefinitions, inclusion rules, exclusion rules, and one positive example for each code.Blank ENA decision recordA reusable worksheet for recording the question, units, conversations, codes, window, and claim boundary.

  1. Step 1

    Name the view before interpreting it

    Record whether you are looking at individual points, group means, one weighted network, or a subtraction network. Note the axis labels, variance shown by the interface, scale settings, and whether node positions are shared across the compared networks. A thicker line in one network is not automatically the same object as a colored line in a difference graph.

    CheckpointYour notes identify the exact plot, groups, settings, and visual encoding under discussion.

  2. Step 2

    Describe the pattern without explaining it

    Start with a literal observation: the scaffolded mean network has a relatively stronger Evidence-Revision connection under this model. Avoid language such as caused, improved, or learned. Description protects the boundary between what the model displays and why the pattern may exist.

    CheckpointThe first sentence reports a modeled relationship and contains no causal verb.

  3. Step 3

    Inspect unit-level variation

    Check whether the group pattern is broadly distributed or driven by one or two units. Compare group means with individual points and networks. If the interface provides uncertainty or statistical comparison tools, record the test, grouping, sample size, and result rather than treating visual separation as significance.

    CheckpointYou can state whether the visible mean difference is consistent across units or concentrated in a few teams.

  4. Step 4

    Return to the coded evidence

    Use Data View or the source CSV to inspect rows contributing to Evidence and Revision within the selected conversation window. Read the utterances before and around each coded row. Verify that the excerpts support the coding definitions and that the modeled proximity is meaningful in the activity context.

    CheckpointAt least two excerpts are linked to the edge, with unit, conversation, and row identifiers retained.

  5. Step 5

    Test plausible alternatives

    Ask whether the pattern changes under a defensible alternative window, after reviewing a borderline code, or when one influential unit is removed. Also inspect code frequencies: ENA focuses on relations, but a rare or unevenly applied code can still shape the network. Treat sensitivity checks as part of interpretation rather than as a search for the preferred picture.

    CheckpointYour record includes at least one justified sensitivity check and its outcome.

  6. Step 6

    Write a layered claim

    Write three linked statements: what the ENA model shows, what the underlying excerpts suggest, and what the design cannot determine. For this tutorial, an appropriate conclusion is that the synthetic scaffolded records contain a stronger modeled connection between evidence and revision, illustrated by excerpts where teams revise a strategy in response to student work. The dataset cannot show that a real scaffold would cause better reasoning or learning.

    CheckpointThe final paragraph separates modeled pattern, contextual interpretation, and limitation.

Continue the pathway